The role of perceptual and conceptual similarity in early word – and referent processing in a preferential looking task
Introduction
It has been commonly reported in diary reports (Rescorla, 1979) and confirmed in empirical research (Thomson and Chapman, 1977; Fremgen and Fay, 1980; Gelman and Naigles, 1995) that children aged one to two years generalise word meanings to a wider collection of objects than adults would consider appropriate. For example, a child may refer to all four-legged animals using the label ‘doggie’. This phenomenon became known as ‘overextension error’. Children’s generalisations of words have been typically regarded by adults as inappropriate and an indication of early misconceptions. However, more recently it has been suggested that children’s word generalisation in comprehension tasks may reflect what information children are accessing from their semantic categories. In this introduction, three main topics will be discussed.
Theories and methods for the study of overextension and early word comprehension
Barrett (1977) provided a brief overview and critique of the early theoretical approaches to word comprehension and overextension in young children. The focus of these theoretical approaches was on how children associate referent features with newly learnt words. One approach was offered by the Semantic Feature hypothesis (SFH) stated by Clark et al (1973, cited in Barrett, 1977). This states that the meaning of a word is comprised of the semantic features of its referent. An alternative view was offered by the Functional Core Hypothesis (FCH) stated by Nelson et al (1974, cited in Barrett, 1977). According to this hypothesis, children apply words to referents on the basis of their functional attributes. Both the SFH and FCH theories assume that as children extract more referent features associated with words, they will be able to limit their extension. However Barrett (1977) argues that it is not clear from these accounts how children differentiate between referents within a category.
Barrett (1977) proposed a new contrastive theory which stated that the meaning of a word can only be acquired by differentiating between positive and negative word referents. According to Barrett (1977)’s contrastive theory, children are able to form semantic categories using referent features as assumed by the SFH and FCH theories. However the contrast theory also states that within a semantic category there may be a network of subdivisions which serve to distinguish between words. When a new word is learnt, it must first satisfy an entry condition for an existing category in the child’s lexicon; i.e. the word ‘cat’ would be allocated to the semantic category for animals. Then the word must surpass a series of networks containing contrasting options. For example, within the semantic category for animals there may be subdivisions for animals with and without body hair. Using these systems children may overextend words to a smaller group of more appropriate examples. Barrett (1977) predicted that new words will not be overextended to objects which have already been given more appropriate labels via semantic networks.
Thomson and Chapman (1977) were the first to address the status of two-year-olds’ overextensions in an empirical study. The aim of their research was to test two interpretations of overextension data. One interpretation was that word meanings are represented by referent features which determine how they are used as suggested by the SFH theory. From this point of view, it is suggested that words overextended in production should also be overextended in comprehension. An alternative view is that children may be able to differentiate between objects which they label incorrectly. Therefore, overextended word-use would be attributed to retrieval errors. Thomson and Chapman (1977) investigated these issues by comparing word comprehension and production in five two-year-old children.
Thomson and Chapman’s (1977) experimental procedure involved a number of sessions in which information regarding typical overextensions was collected from parental interviews and interactions with the children. The sessions with the children involved two tasks which were used to examine the children’s word-use and comprehension. The children first took part in a production task in which they were presented with a deck of images. The children were asked to identify who or what was in each image. The second task assessed each child’s comprehension of four words which had been overextended in the production task. This involved a series of trials in which the children were presented with pairs of images on a presentation board. During each trial the children were asked to point to the image that matched the referent of the word they heard.
The main findings of Thomson and Chapman’s (1977) study showed that some words overextended in production were not overextended in comprehension while others were. This implies that overextensions are not based purely on either the children’s underlying word meanings or retrieval errors. Rather these findings seem to indicate the existence of two forms of overextension in production and comprehension. Overextensions of words in only production were attributed to retrieval errors. The children were aware of appropriate examples for the target words in the comprehension trials but when asked to name these objects in the production task, they perhaps could not recall the appropriate words. The comprehension task assessed the children’s recognition of appropriate objects associated with the target words as opposed to their word-use. Overextensions in comprehension were therefore attributed to the children’s underlying representations of word meanings.
Fremgen and Fay (1980) challenged Thomson and Chapman’s (1977) findings and argued that the overextensions reported in their study were elicited by repeated testing. Fremgen and Fay (1980) attempted to eliminate this problem by testing each child’s comprehension only once. Fremgen and Fay (1980) assessed word comprehension and production for 16 children in two tasks. The first task examined the children’s overextended word use. The children were presented with a series of images and asked to name them. The second task assessed the children’s word comprehension. The children were presented with four images spread out in front of them. The array consisted of an appropriate image, an inappropriate image and two distracters. While the children looked at the images they were asked to find an appropriate image which matched the referents of the words they heard in a series of trials (i.e. dog for the word ‘dog’). This was followed by a second set of trials in which the children were asked to find the images of objects which had been incorrectly labelled in the production task. For instance, a child who labelled a cat with the word ‘dog’ would be asked to find ‘cat’ in the comprehension trials. The results of the study showed that the children did overextend in the production task but not in any of the comprehension trials. Fremgen and Fay (1980) concluded that children are always aware of the correct referents for overextended words and that all overextensions merely reflect performance errors.
An alternative approach to studying overextensions in children’s early language development was offered by Behrend (1988). Behrend (1988) argued that previous research has focused too much on overextensions in production. He suggests that because previous research (Thomson and Chapman 1976) has revealed differences between overextensions in production and comprehension it would be interesting to look at whether overextensions occur at an earlier stage of word comprehension preceding production. If the infants in Behrend’s study did show overextensions in a comprehension task then this would confirm that independent overextensions in comprehension do exist and cannot be attributed to retrieval errors. Therefore Behrend’s sample consisted of infants aged 13 months.
Behrend (1988) used a methodological approach known as the ‘signal detection paradigm’ to test the infants’ early language comprehension. Before the experiment, parents were asked to select words from a list, two of which were familiar and two which were unfamiliar to their child. The child was seated in a high chair in front of a screen with curtains. Behind the curtains, referents for the familiar and unfamiliar words along with two distracters were placed on four shelves. During each trial, parents would say each word to their child whilst the child saw the referents. A second experiment was conducted which followed a similar procedure to the first experiment with the following exceptions. The unknown words had perceptually similar referents to the familiar words. For half of the trials only the familiar word referents and three distracters were presented. During the remaining trials only the unknown word referents and three distracters were presented.
The main findings of Behrend’s (1988) study were that in the first experiment the infants looked longer at familiar word referents when they heard familiar words. Therefore, this indicated that 13- month-old infants could comprehend the meanings of familiar words. However in the second experiment infants also looked longer at the unknown referents when they heard familiar words. Therefore, this indicates that 13-month-old infants will also overextend words when presented with perceptually similar referents. Behrend (1988) proposed two possible interpretations of these findings. One interpretation was that during the overextension trials, the children were aware that the appropriate referent was unavailable and instead, incorrectly, looked at a related referent. Alternatively, these findings may reflect how children represent word meanings and what information they are accessing from their basic semantic categories. Behrend (1988) concluded that overextensions in comprehension do exist but could not be attributed to production errors.
Gelman and Naigles (1995) investigated two explanations of children’s overextensions drawn from previous research. One argument is that overextensions reflect the development of children’s semantic categories. Previous theoretical notions (Barrett, 1977) have suggested that overextensions may reflect how children represent word meanings. Children’s semantic categories may have fewer lexical entries than that of an adult with unclear boundaries based on referent features. Behrend (1988), in his comprehension trials with pre-verbal infants, found that overextensions reported in comprehension can occur independently and much earlier; before production errors. An alternative argument is that overextensions merely reflect errors in language production as suggested by Fremgen and Fay (1980). Gelman and Naigles (1995) argued that if overextension errors are merely production errors, comprehension should not be affected.
Gelman and Naigles (1995) tested their hypothesis with infants aged one to two years in a preferential looking experiment. Before the experiment, the infants took part in a language production task in which they were asked to label three animal puppets. The infants were then allocated to one of two groups according to whether they labelled the puppets correctly (non-overextended) or incorrectly (overextended). Both groups then took part in a preferential looking experiment. The children were seated on their mother’s lap and saw two videos of the same three animal puppets used in the production task simultaneously on two screens. Whilst observing the two videos the children heard an auditory recording of one of the three puppet’s names.
The results of Gelman and Naigles’s (1995) study showed that, consistent with previous research (Fremgen and Fay, 1988), both groups of infants were aware of the correct referents for all words, even the ones they overextended. Overextensions reported in the production task were attributed to retrieval errors. Overextensions reported in the comprehension trials, however. did not match the pattern of overextensions reported in the production task. For example, when the children heard the word ‘cow’ and were only presented with dog and cat puppets they would look longer at the dog. In the production task the children did not overextend the word ‘cow’, but instead the word ‘dog’. Gelman and Naigles (1995) proposed that there are two independent types of overextension in language production and comprehension which seem to arise from different sources. Overextensions in production seem to be caused by difficulties with retrieving appropriate words. Overextensions in comprehension seem to reflect children’s processing of similarities of features stored in their semantic categories.
Johnson and Huettig (2010) investigated what information children are accessing when searching for absent referents in comprehension trials such as looking for a cow when only a dog and cat are present. Previous research has indicated that hearing a familiar word mediates a shift in visual attention towards an associated image. According to Johnson and Huettig (2010), when the actual referent is absent, children appear to activate various attributes associated with a word and use this to search for the appropriate referent. However it is unclear in these trials what information children are accessing because the referents are often similar across several domains. For example, the puppets of Gelman and Naigles (1995) shared similar perceptual features and fell under the same taxonomical category.
In their present study, Johnson and Huettig (2010) investigated whether infants aged 36 months access and use individual perceptual attributes related to colour in word-recognition tasks. In their study, infants took part in a preferential looking paradigm consisting of three types of trial; target trials, unrelated distracter trials and colour-matched trials. In the target trials, the infants were asked to search for one of two familiar images presented. In the unrelated trials the infants were presented with two images matched for all perceptual features except colour and which were completely unrelated to the spoken words. The colour-match trials were very similar to the unrelated trials with one exception. The spoken words had a particular colour associated with them which matched one of the two images. It was predicted that if infants do access and process colour attributes, they would look longer at the colour-matched image. As predicted, the infants looked significantly longer at the colour-matched image. Therefore it was concluded that infants do access specific attributes associated with words which mediates their visual processing of referents. However, there were questions which still needed to be addressed. Johnson and Huettig (2010) suggest that children may attend to some attributes more than others and that this weighing of attributes may vary across children of different ages.
The Shape Bias and the role of shape in early word learning
Early theories such as the SFH and the FCH (Barrett 1977) introduced the notion that children learn new words by attending to the similarities between referent features. As a result,newly learnt words are often overextended to referents which share similar features such as shape, texture and function. This is confirmed in numerous empirical studies. For example, Rescorla (1979) collected vocabulary data from parental diary reports of six children over a period of six months. The largest proportion of extensions reported was perceptually based. The majority of these extensions were based on visual characteristics. Thomson and Chapman (1977) carried out addition analyses of the images used in their experiments. The findings showed that referents which were incorrectly given the same name shared similar perceptual features.
It has been suggested that infants tend to apply novel words to referents with similar perceptual features, mainly related to shape, because they believe that objects with a similar shape also have the same name. This phenomenon is known as the ‘shape bias’ and has been the focus of much research. There are currently two theoretical approaches to the shape bias supported by empirical research; the associative learning account (ALA) (Smith, 2000) and the ‘shape-as-cue’ account (Markson, Diesendruck and Bloom, 2008).
The ALA claims that the shape bias emerges during associative word learning. It is proposed that, over time, infants will form associations between words and groups of appropriate objects with a similar shape via repeated exposure and demonstration. As a result, the infants will automatically attend more to shape when applying new words to new objects. The strongest evidence for this comes from training studies in which children with limited vocabularies are taught word categories organised predominantly by shape (Smith 2000). Their findings confirmed that, consistent with the ALA, the children applied known words to object categories on the basis of shape after training.
An alternative theory to the shape bias, the ‘shape-as-cue’ account (Markson, Diesendrunk and Bloom, 2008) states that young children understand how some words refer to categories of objects and that shape can be used as cue for determining appropriate object categories prior to word learning. This account emphasises children’s existing categorical knowledge and suggests that training effects, reported in studies, merely demonstrate how training reinforces children’s innate ability to generalise words on the basis of shape.
Research has revealed that infants’ word overextensions or over-generalisations can be based on other taxonomical similarities. Infants will also apply words to objects which share several categorical or perceptual features and thus fall into the same taxonomical categories. Rescorla (1979) identified overextensions which she called categorical over exclusions, defined as the application of words to referents which fall under similar semantic categories such as ‘people’. Both the ‘shape as cue’ and ALA acknowledge that children can develop attention biases other than shape, but offer alternative explanations of how they are controlled. The ‘shape as cue’ account suggests that shape may be over-ridden by more salient cues such as function or texture. The ALA suggests that shape may be used in combination with other perceptual or taxonomical attention biases, (Smith, 2000).
There is the possibility of a developmental shift in word-learning biases from shape-based to taxonomy-based. Jones, Smith and Landau (1991, cited in Smith, 2000) provide some evidence for this in their study in which they examined the effect of perceptual cues on children’s word generalisations. The children were shown objects with or without eyes which had a similar shape and/ or texture as objects labelled by novel words. Two-year-old children generalised the novel words to objects purely on the basis of shape. Three-year-olds, however, were more selective when generalising the novel words to eyed- and eyeless objects. It is not yet clear whether children younger than two years show attention biases when generalising words or whether this developmental trend is evident at a younger age. A further issue is that perceptual and categorical similarities are often confounded. Objects which look the same often fall within the same taxonomical category. In the next section, two papers which explore young infants’ sensitivity to perceptual and categorical similarity will be discussed.
References
Barrett, M. D. (1977). Lexical Development and Overextension in child language. Journal of Child Language, 5, 205-219.
Behrend, D. A. (1988). Overextensions in early language comprehension: evidence from a signal detection approach. Journal of Child Language, 15, 63-75.
Fremgen A., & Fay D. (1980). Overextensions in production and comprehension a methodological clarification. Journal of Child Language, 7, 205-211.
Gelman, S. A., & Niagles, L.G. (1995). Overextensions in comprehension and production revisited: preferential looking in a study of dog, cat and cow. Journal of Child Language, 22, 19-64.
Johnson, E.K. & Huettig, F. (2010 in press). Eye movements during language-mediated visual search reveal a strong link between overt visual attention and lexical processing in 36-months-olds. Psychological Research. 1-28.
Markson, L. Diesendrunk, G. and Bloom, P. (2008). Shape Bias Special Section: the shape of thought. Developmental Science,11, 204-208.
Rescorla, L. A. (1979). Overextension in early language development. Journal of Child Language, 7, 321-335.
Smith, L.B. (2000). Learning How to Become a Word Learner: An Associative Crane: In (Golinkoff, R. Hirsh-Pasek, K. Bloom, L, Smith, L. B., Woodward A.L & Akhtar, N. et al.) Becoming a word learner: A Debate on Lexical Acquisition, New York: Oxford University Press.
Thompson J. R. & Chapman, R. S. (1977). Who is Daddy revisited: the status of two-year olds’ over-extended words in use and comprehension. Journal of Child Language, 4 359-375.